Parallel Chinese-English Entities, Relations and Events Corpora

Parallel Chinese-English Entities, Relations and Events Corpora Justin Mott, Zhiyi Song, Ann Bies, Stephanie Strassel Linguistic Data Consortium, University of Pennsylvania 3600 Market Street, Suite 810, Philadelphia, PA 19104 USA E-mail: {jmott,zhiyi,bies,strassel}@ldc.upenn .edu Abstract This paper introduces the parallel Chinese-English Entities, Relations and Events (ERE) corpora developed by Linguistic Data Consortium under the DARPA Deep Exploration and Filtering of Text (DEFT) Program. Original Chinese newswire and discussion forum documents are annotated for two versions of the ERE task. The texts are manually translated into English and then annotated for the same ERE tasks on the English translation, resulting in a rich parallel resource that has utility for performers within the DEFT program, for participants in NIST’s Knowledge Base Population evaluations, and for cross-language projection research more generally. Keywords: language resources, information extraction, entities, relations, events 1. Introduction This paper introduces the parallel Chinese-English Entities, Relations and Events (ERE) corpora developed at the Linguistic Data Consortium (LDC) as a part of DARPA’s Deep Exploration and Filtering of Text (DEFT) program. The DEFT program seeks to improve state-of-the-art capabilities in automated deep natural language processing, with a particular focus on technologies dealing with inference, causal relationships, and anomaly detection across several languages (DARPA 2012). Given the large number and variety of approaches to algorithm development within DEFT, we set out to define an annotation task that would be supportive of multiple research directions and technology evaluations, and that would provide a useful foundation for follow-on DEFT annotation tasks like entailment, inference and belief/sentiment. The resulting Entities, Relations and Events annotation task has evolved over the course of the DEFT program, from a fairly lightweight treatment of entities, relations and events in text, to a richer representation of phenomena of interest to the program (Song et al. 2015). ERE corpora are used by DEFT performers as a general resource, and also serve as training data for several tracks within the Text Analysis Conference Knowledge Base Population (TAC KBP) evaluation series, which is open to non-DEFT participants conducted by the National Institute of Standards and Technology (NIST). TAC KBP aims to develop and evaluate technologies for building and populating knowledge bases from unstructured texts (NIST 2015). The ERE corpora provide a training resource for component evaluation tasks such as Entity Detection and Linking and Event Argument Linking. In keeping with the overarching goals of the DEFT Program, TAC KBP increasingly focuses on extracting information from multilingual resources (Ji 2010; NIST 2015), and the parallel ERE data described in this paper are particularly useful in this context. The parallel ERE corpora have relevance beyond the specific objectives of DEFT and TAC KBP. Cross- language projection is an important means to bootstrap the transfer of annotation across multiple languages, and has been applied to many annotation schemes, both with bitext (Yarowsky et al., 2001; Wang and Manning, 2014; Zitouni, 2008; Ehrmann et al., 2011) and without bitext (Zirikly 2014). The creation of independently annotated parallel data sets, such as the ones described here, can serve as …

Content truncated to 500 words